Merge pull request #44 from louis030195/d-#32

#32

- use tokio in screenpipe-vision
- use n parallel tasks (should be same behaviour than before)
- (new) skip frame if cpu overloaded for OCR (configurable)
- can now turn on/off devices from screenpipe-audio using API
- can now turn on/off vision using API
- disconnecting audio device in use stop listening. User need to make API request to enable any new device (could be improved later)

.cargo/config.toml

@@ -0,0 +1,3 @@
+[build]
+# for console_subscriber
+rustflags = ["--cfg", "tokio_unstable"]

Cargo.toml

@@ -18,3 +18,5 @@ candle-nn = { package = "candle-nn", version = "0.6.0" }
 candle-transformers = { package = "candle-transformers", version = "0.6.0" }
 tokenizers = "0.19.1"
 tracing = "0.1.37"
+tokio = { version = "1.15", features = ["full", "tracing"] }
+

screenpipe-audio/Cargo.toml

@@ -59,9 +59,13 @@ crossbeam = "0.8"
 # Bytes
 bytemuck = "1.16.1"
 
+# Async
+tokio = { workspace = true }
+
 [dev-dependencies]
 tempfile = "3.3.0"
 infer = "0.15"
+tokio = { version = "1.0", features = ["full"] }
 
 [features]
 metal = ["candle/metal", "candle-nn/metal", "candle-transformers/metal"]

screenpipe-audio/src/bin/screenpipe-audio.rs

@@ -5,10 +5,12 @@ use screenpipe_audio::create_whisper_channel;
 use screenpipe_audio::default_input_device;
 use screenpipe_audio::default_output_device;
 use screenpipe_audio::list_audio_devices;
-use screenpipe_audio::parse_device_spec;
+use screenpipe_audio::parse_audio_device;
 use screenpipe_audio::record_and_transcribe;
 use screenpipe_audio::AudioDevice;
 use std::path::PathBuf;
+use std::sync::atomic::AtomicBool;
+use std::sync::Arc;
 use std::thread;
 use std::time::Duration;
 
@@ -33,6 +35,8 @@ fn print_devices(devices: &[AudioDevice]) {
     }
 }
 
+// TODO - kinda bad cli here
+
 fn main() -> Result<()> {
     use env_logger::Builder;
     use log::LevelFilter;
@@ -56,27 +60,41 @@ fn main() -> Result<()> {
     } else {
         args.audio_device
             .iter()
-            .map(|d| parse_device_spec(d))
+            .map(|d| parse_audio_device(d))
             .collect::<Result<Vec<_>>>()?
     };
 
     if devices.is_empty() {
         return Err(anyhow!("No audio input devices found"));
     }
 
-    let chunk_duration = Duration::from_secs(30);
+    // delete .mp3 files (output*.mp3)
+    std::fs::remove_file("output_0.mp3").unwrap_or_default();
+    std::fs::remove_file("output_1.mp3").unwrap_or_default();
+
+    let chunk_duration = Duration::from_secs(5);
     let output_path = PathBuf::from("output.mp3");
     let (whisper_sender, whisper_receiver) = create_whisper_channel()?;
-
     // Spawn threads for each device
     let recording_threads: Vec<_> = devices
         .into_iter()
         .enumerate()
         .map(|(i, device)| {
             let whisper_sender = whisper_sender.clone();
             let output_path = output_path.with_file_name(format!("output_{}.mp3", i));
+            let device_control = Arc::new(AtomicBool::new(true));
+            let device_clone = device.clone();
+
             thread::spawn(move || {
-                record_and_transcribe(&device, chunk_duration, output_path, whisper_sender)
+                let device_control_clone = Arc::clone(&device_control);
+
+                record_and_transcribe(
+                    &device_clone,
+                    chunk_duration,
+                    output_path,
+                    whisper_sender,
+                    device_control_clone,
+                )
             })
         })
         .collect();